OpenClaude Studio

OpenClaude Studio is a read-only companion dashboard for OpenClaude. Run a small local API on your machine, open the hosted web app, and inspect projects, sessions, plans, tasks, providers, diagnostics, usage, and debug logs without editing local OpenClaude data.

The project is intentionally scoped around a small read-only foundation: a hosted or local web app talks to a local server running on your machine. The server reads OpenClaude files from disk, redacts likely secrets, and exposes only read-only HTTP endpoints.

OpenClaude Studio is an independent community companion dashboard for OpenClaude. It is not affiliated with, endorsed by, or sponsored by Anthropic.

Why OpenClaude Studio?

OpenClaude keeps useful local context in files that are awkward to inspect by hand. Studio gives that data a safer dashboard surface:

• See which OpenClaude projects were active recently.
• Inspect sessions, assistant messages, tool calls, tool results, errors, changed files, usage, and cost.
• Review linked plans, linked tasks, and checklist progress without opening raw task files.
• Search project-scoped debug logs without opening raw log files.
• Review provider profiles, validation diagnostics, and safe templates while keeping secret fields redacted.
• Diagnose missing, malformed, or unavailable local data.

Current Scope

The current MVP line includes:

• Project selector backed by ~/.openclaude.json plus bounded transcript metadata from ~/.openclaude/projects/
• Project overview with recent sessions, usage chart, log issue counts, and provider status
• Session list and rich session inspector
• Conversation timeline with user messages, assistant messages, tool calls, tool results, and errors
• Read-only Session Change Review with bounded, redacted diffs from file-history backups and current project files
• Changed files, token usage, cost summary, linked plans, linked tasks, and file-history context for sessions
• Plans & Tasks view with checklist progress, linked sessions, task status groups, and detail panes
• Read-only provider profile management with active profile inspection, validation diagnostics, and safe templates
• Project-scoped diagnostics
• Project-scoped debug log viewing, filtering, search, virtualized scrolling, and copy-to-clipboard for log messages
• Dark and light themes
• Visible local API URL control for custom local server ports
• Read-only local API with no write endpoints

It does not currently edit OpenClaude settings, provider profiles, sessions, logs, project files, tasks, plans, or file-history data.

Screenshots

Screenshots use synthetic fixture data only.

How It Works

Browser UI
  |
  | HTTP JSON
  v
Local server on 127.0.0.1:43110
  |
  | read-only file access
  v
~/.openclaude.json
~/.openclaude/projects/
~/.openclaude/plans/
~/.openclaude/tasks/
~/.openclaude/file-history/
~/.openclaude/debug/

The web UI can run locally during development or be hosted as static assets. The server should run on the same machine as OpenClaude because it reads local OpenClaude files.

Privacy At A Glance

• The local API binds to 127.0.0.1 by default.
• The API is read-only and has no write endpoints in the current MVP line.
• The hosted web app runs in your browser and talks to your local server.
• The app code does not intentionally collect telemetry; review your own hosting configuration if you deploy a fork.
• Project discovery may show project paths found in OpenClaude transcript metadata, including historical paths that are no longer present on disk.
• Redaction is defense in depth, not a guarantee for every possible secret format.
• Review logs, screenshots, and recordings before sharing them publicly.

Requirements

• Node.js 22 or newer
• npm
• OpenClaude installed and used at least once, so local config/session files exist

The supported runtime floor is Node.js 22. CI and release jobs use Node.js 22 so local development, validation, and publishing exercise the same supported baseline. The repository includes .nvmrc for contributors who use nvm-compatible tooling.

Quick Start

1. Start the local read-only API:

npx openclaude-studio

2. Open the hosted dashboard:

https://openclaude-studio.pages.dev/

3. Keep the terminal running while you use the dashboard.

The browser UI connects to http://127.0.0.1:43110 by default. If the local server is not running, the app will show the command above and the API URL it expected to reach. See Troubleshooting for common setup and connection issues.

Local Development

Install dependencies:

npm install

Run the local server and web app:

npm run dev

Open the web UI at:

http://127.0.0.1:5173

The local API listens at:

http://127.0.0.1:43110

Production-Style Local Run

Build all workspaces:

npm run build

Start the local API:

npm run start -w openclaude-studio

By default, the server binds to 127.0.0.1 and listens on port 43110.

Configuration

Variable	Default	Description
OPENCLAUDE_STUDIO_HOST	127.0.0.1	Host for the local API server. Keep this on loopback unless you provide your own trusted access control.
OPENCLAUDE_STUDIO_PORT	43110	Port for the local API server.
OPENCLAUDE_STUDIO_ALLOWED_ORIGINS	official hosted app plus loopback browser origins	Comma-separated additional hosted web origins allowed to call the local API.
OPENCLAUDE_STUDIO_TOKEN	unset	Optional API token for custom callers or deployments with their own access flow. The bundled web UI does not prompt for tokens.
CLAUDE_CONFIG_DIR	~/.openclaude	OpenClaude config directory override, useful for testing alternate local data roots.

The official hosted app at https://openclaude-studio.pages.dev is allowed by default. If you host the web UI somewhere else, add that origin:

OPENCLAUDE_STUDIO_ALLOWED_ORIGINS=https://studio.example.com npm run start -w openclaude-studio

Development

Useful commands:

npm run lint
npm test
npm run build
npm run smoke:package
npm run test:e2e

Workspace layout:

• apps/server: Fastify local API and CLI binary
• apps/web: Vite, React, Tailwind CSS dashboard
• packages/shared: API response types shared by server and web
• tests/e2e: Playwright coverage for the integrated app

Additional documentation:

• Architecture
• Local server
• Troubleshooting
• Deployment
• Privacy and redaction

Safety Model

OpenClaude Studio is designed to be conservative by default:

• The current API is read-only.
• The server binds to loopback by default.
• Browser origins are restricted to loopback, the official hosted app, and any origins explicitly configured by the user.
• File reads are bounded.
• Symlink traversal is avoided for sensitive local file reads.
• Provider URLs, auth fields, bearer tokens, common API key formats, and log messages are redacted where possible.

Redaction is defense in depth, not a guarantee for every possible secret format. Avoid sharing screenshots or logs without reviewing them.

Roadmap

The current release focuses on a useful read-only foundation. Future work will be prioritized by real user feedback. Valuable next areas include:

Shipped in the MVP line:

• Hosted web dashboard
• Installable local API package through npm
• Project selector and project overview
• Session summaries for the selected project
• Rich session timeline with transcript, tool call, file change, usage, and error details
• Session Change Review with read-only per-session diffs, backup status, diagnostics, and risk flags
• Plans and tasks views linked back to sessions
• File history and backup context for selected sessions
• Active provider inspection with secret fields redacted
• Project-scoped diagnostics
• Debug log viewing, filtering, search, virtualized scrolling, and log-message copy
• Dark and light themes
• Custom local API URL setting for non-default ports
• Provider profile management with safe templates and validation
• Conservative read-only local API

Planned and open for discussion:

• Global project search across sessions, logs, config, prompt assets, plans, and tasks
• Broader file history and backup inventory across selected projects
• Config source explorer for user settings, project settings, local settings, and managed config
• Prompt asset inventory for instructions, agents, commands, workflows, and output styles
• Hooks and permissions diagnostics
• Live log streaming with pause, filtering, and retention controls
• Optional write workflows after the write model, review UX, backups, and security boundaries are designed explicitly

Ideas, bug reports, and focused pull requests are welcome. If you propose a write-capable feature, please include the expected safety model and rollback behavior.

Contributing

Please read CONTRIBUTING.md before opening a pull request. Keep changes focused, include tests for behavior changes, and avoid committing local data, generated output, logs, secrets, or machine-specific paths.

Security

Please read SECURITY.md for the local data access model and vulnerability reporting guidance.

License

MIT. See LICENSE.